ai-dynamo/grove — Kubernetes enhancements for Network Topology Aware Gang Scheduling & Autoscaling

Description and use

The **ai-dynamo/grove** project extends Kubernetes with topology‑aware gang scheduling and autoscaling, enabling AI/ML workloads to be placed and scaled based on the underlying network fabric. By exposing network topology information to the scheduler, Grove ensures that pods that need to communicate frequently are co‑located on low‑latency nodes, while its autoscaling component dynamically adjusts replica counts to match both compute demand and network constraints. This results in more efficient resource utilization and reduced training times for distributed AI workloads.

Problem

ai-dynamo/grove helps make deployment and operations more repeatable.

Outcome

Medium: useful for prototypes or internal workflows, with dependency and maintenance checks before production.

Workflow

• standardize deployment
• automate operations
• improve platform reliability

Evidence

Kubernetes enhancements for Network Topology Aware Gang Scheduling & Autoscaling

Source